bump accelerate to 0.34.2 (#1901)

* bump accelerate

* add fixture to predownload the test model

* change fixture

.github/workflows/multi-gpu-e2e.yml

@@ -1,6 +1,9 @@
 name: docker-multigpu-tests-biweekly
 
 on:
+ pull_request:
+ paths:
+ - 'tests/e2e/multigpu/*.py'
  workflow_dispatch:
  schedule:
  - cron: '0 0 * * 1,4' # Runs at 00:00 UTC every monday & thursday

requirements.txt

@@ -4,7 +4,7 @@ peft==0.12.0
 transformers==4.44.2
 tokenizers>=0.19.1
 bitsandbytes==0.43.3
-accelerate==0.34.0
+accelerate==0.34.2
 datasets==2.20.0
 deepspeed==0.14.4
 pydantic==2.6.3

tests/e2e/multigpu/test_llama.py

@@ -10,6 +10,7 @@
 import pytest
 import yaml
 from accelerate.test_utils import execute_subprocess_async
+from huggingface_hub import snapshot_download
 
 from axolotl.utils.dict import DictDefault
 
@@ -19,6 +20,12 @@
 os.environ["WANDB_DISABLED"] = "true"
 
 
+@pytest.fixture(scope="session", autouse=True)
+def download_model():
+ # download the model
+ snapshot_download("TinyLlama/TinyLlama_v1.1")
+
+
 class TestMultiGPULlama(unittest.TestCase):
  """
  Test case for Llama models using LoRA